Search

Funding Information: We acknowledge the computational resources provided by the Aalto Science-IT project and the Finnish Functional Genomics Centre and Biocenter Finland. Funding Information: This work was supported by the Academy of Finland (292660, 311584, 335436). The funding body played no role in the design of the study, the collection, analysis, interpretation of data, or in writing the manuscript. Publisher Copyright: © 2022, The Author(s). Background: DNA methylation is commonly measured using bisulfite sequencing (BS-seq). The quality of a BS-seq library is measured by its bisulfite conversion efficiency. Libraries with low conversion rates are typically excluded from analysis resulting in reduced coverage and increased costs. Results: We have developed a probabilistic method and software, LuxRep, that implements a general linear model and simultaneously accounts for technical replicates (libraries from the same biological sample) from different bisulfite-converted DNA libraries. Using simulations and actual DNA methylation data, we show that including technical replicates with low bisulfite conversion rates generates more accurate estimates of methylation levels and differentially methylated sites. Moreover, using variational inference speeds up computation time necessary for whole genome analysis. Conclusions: In this work we show that taking into account technical replicates (i.e. libraries) of BS-seq data of varying bisulfite conversion rates, with their corresponding experimental parameters, improves methylation level estimation and differential methylation detection.

Additional file 1: Additional figures containing: (i) Demonstrating differences in technical parameters, (ii) Testing different hyperparameters for sequencing error and bisulfite conversion rates, (iii) Testing different $$\sigma_B^2$$ �� B 2 for methylation level estimation, (iv) Choosing parameters for variational inference, (v) Comparing full and reduced models in methylation level estimation, and (vi) True positive rates of differential methylation.

Supplementary Figure 1. Performance evaluation of the six copy number variation (CNV) algorithms using the simulated data with the loose criteria: at least 60% overlap between the inferred and ground truth CNV segments and inclusion of ≥ 0.5Mbp CNV segments. A) True positive rate(TPR), B) False discovery rate (FDR), and C) F1 score of the CNV detections achieved by the different tools when the read coverage is varied. The data points are based on the window size comparison results (Supplementary Figures 2-6), from which we selected the window settings that provided the highest F1 scores by the algorithms at each read coverage. Error bars denote the standard error of the results produced with 20 different random subsets. Supplementary Figure 2. Analysis of how the window size affects the performance of CNVnator at different read coverages with simulated data. A) True positive rate (TPR), B) False discovery rate (FDR), and C) F1 score. The hard criteria (minimum overlap of 0.8 and no filtering by size) were used in the analysis. Supplementary Figure 3. Analysis of how the window size affects the performance of BICseq2 at different read coverages with simulated data. A) True positive rate (TPR), B) False discovery rate (FDR), and C) F1 score. Default window size is 0.1 kbp. The hard criteria (minimum overlap of 0.8 and no filtering by size) were used in the analysis. Supplementary Figure 4. Analysis of how the window size affects the performance of FREEC at different read coverages with simulated data. A) True positive rate (TPR), B) False discovery rate (FDR), and C) F1 score. The coefficient of variation of 0.05 is the default value of the built-in method of FREEC for selecting the window size based on the coverage. The hard criteria (minimum overlap of 0.8 and no filtering by size) were used in the analysis. Supplementary Figure 5. Analysis of how the window size affects the performance of HMMcopy at different read coverages with simulated data. A) True positive rate (TPR), B) False discovery rate (FDR), and C) F1 score. The hard criteria (minimum overlap of 0.8 and no filtering by size) were used in the analysis. Supplementary Figure 6. Analysis of how the window size affects the performance of QDNAseq at different read coverages with simulated data. A) True positive rate (TPR), B) False discovery rate (FDR), and C) F1 score. The hard criteria (minimum overlap of 0.8 and no filtering by size) were used in the analysis. Supplementary Figure 7. Visualization of the CNVs detected in the H9-AB-p116 dataset using the six algorithms along with the array-based benchmark CNV segments in the respective chromosomal locations. Deletions are marked in red and gains in blue. The bottom part of the visualization depicts the depth of read coverage at each 50 kbp window. The visualization includes every CNV found with each tool using the window size that yielded the best performance for the simulated data at coverage of 0.1x (see Supplementary Figures. 2-6). Supplementary Figure 8. Visualization of the CNVs detected in the H9-AB-p113 dataset using the six algorithms along with the array-based benchmark CNV segments in the respective chromosomal locations. Deletions are marked in red and gains in blue. The bottom part of the visualization depicts the depth of read coverage at each 50 kbp window. The visualization includes every CNV found with each tool using the window size that yielded the best performance for the simulated data at coverage of 0.1x (see Supplementary Figures. 2-6). Supplementary Figure 9. Visualization of the CNVs detected in the H9-p38 dataset using the six algorithms along with the array-based benchmark CNV segments in the respective chromosomal locations. Deletions are marked in red and gains in blue. The bottom part of the visualization depicts the depth of read coverage at each 50 kbp window. The visualization includes every CNV found with each tool using the window size that yielded the best performance for the simulated data at coverage of 0.1x (see Supplementary Figures. 2-6). Supplementary Figure 10. Visualization of the CNVs detected in the H9-p41 dataset using the six algorithms along with the array-based benchmark CNV segments in the respective chromosomal locations. Deletions are marked in red and gains in blue. The bottom part of the visualization depicts the depth of read coverage at each 50 kbp window. The visualization includes every CNV found with each tool using the window size that yielded the best performance for the simulated data at coverage of 0.1x (see Supplementary Figures. 2-6). Supplementary Figure 11. Visualization of the CNVs detected in all the chromosomes in the combined sample H9-AB by the six algorithms along with the array-based benchmark CNV segments in the respective chromosomal locations. Deletions are marked in red and gains in blue. The bottompart of the visualization depicts the depth of read coverage at each 50 kbp window. The visualizationincludes every CNV found with each tool using the window size that yielded the best performancefor the simulated data at coverage of 0.1x (see Supplementary Figures. 2-6). Supplementary Figure 12. Visualization of the CNVs detected in all the chromosomes in the combined sample H9-NO by the six algorithms along with the array-based benchmark CNV segments in the respective chromosomal locations. Deletions are marked in red and gains in blue. The bottom part of the visualization depicts the depth of read coverage at each 50 kbp window. The visualization includes every CNV found with each tool using the window size that yielded the best performance for the simulated data at coverage of 0.1x (see Supplementary Figs. 2-6). Supplementary Figure 13. Visualization of the CNVs detected in all the chromosomes in the combined sample H9-AB-p116 by the six algorithms along with the array-based benchmark CNV segments in the respective chromosomal locations. Deletions are marked in red and gains in blue. The bottom part of the visualization depicts the depth of read coverage at each 50 kbp window. The visualization includes every CNV found with each tool using the window size that yielded the best performance for the simulated data at coverage of 0.1x. Supplementary Figure 14. Visualization of the CNVs detected in all the chromosomes in the combined sample H9-AB-p113 by the six algorithms along with the array-based benchmark CNV segments in the respective chromosomal locations. Deletions are marked in red and gains in blue. The bottom part of the visualization depicts the depth of read coverage at each 50 kbp window. The visualization includes every CNV found with each tool using the window size that yielded the best performance for the simulated data at coverage of 0.1x (see Supplementary Figs. 2-6). Supplementary Figure 15. Visualization of the CNVs detected in all the chromosomes in the combined sample H9-NO-p41 by the six algorithms along with the array-based benchmark CNV segments in the respective chromosomal locations. Deletions are marked in red and gains in blue. The bottom part of the visualization depicts the depth of read coverage at each 50 kbp window. The visualization includes every CNV found with each tool using the window size that yielded the best performance for the simulated data at coverage of 0.1x (see Supplementary Figs. 2-6). Supplementary Figure 16. Visualization of the CNVs detected in all the chromosomes in the combined sample H9-NO-p38 by the six algorithms along with the array-based benchmark CNV segments in the respective chromosomal locations. Deletions are marked in red and gains in blue. The bottom part of the visualization depicts the depth of read coverage at each 50 kbp window. All chromosomes included. Combined sample H9-NO-p38. The visualization includes every CNV found with each tool using the window size that yielded the best performance for the simulated data at coverage of 0.1x (see Supplementary Figs. 2-6). Supplementary Figure 17. Performance evaluation of the six algorithms using the cell line data with the stringent criteria: at least 80% overlap between the inferred and array-validated CNV segments and ≥0.5Mbp CNV length requirement for the detected CNV segment. A,D) True positive rate, B,E) False discovery rate and C,F) F1 score of the CNV detections. The red and blue dots depict the abnormal and normal samples, respectively. With each tool we used the the window size that yielded the best performance for the simulated data at coverage of 0.1x (see Supplementary Figs. 2-6). Supplementary Table 1. Simulated CNV segments that were used to evaluate the tools. Supplementary Table 2. Number of bases and read coverage of the cell line samples for each sample individually and for the combined samples. Supplementary Table 3. Array-based CNV segments ≥500 kbp used to evaluate the tools. Supplementary Table 4. Failure rates for different read coverages with varying window size settings and 20 different down samplings using simulated data.

High-grade serous ovarian cancer is the most common ovarian cancer type. Although the combination of surgery and platinum-taxane chemotherapy provide an effective treatment, drug resistance frequently occurs leading to poor outcome. In order to clarify the molecular mechanisms of drug resistance, the DNA methylation and transcriptomic changes, associated with the development of drug resistance in high-grade serous ovarian cancer, were examined from patient derived malignant ascites cells. In parallel with large-scale transcriptome changes, cisplatin resistance was associated with loss of hypermethylation at several CpG sites primarily localized in the intergenic regions of the genome. The transcriptome and CpG methylome changes in response to cisplatin treatment of both sensitive and resistant cells were minimal, indicating the importance of post-translational mechanisms in regulating death or survival of the cells. The response of resistant cells to high concentrations of cisplatin revealed transcriptomic changes in potential key drivers of drug resistance, such as KLF4. Among the strongest changes was also induction of IL6 in resistant cells and the expression was further increased in response to cisplatin. Also, several other components of IL6 signaling were affected, further supporting previous observations on its importance in malignant transformation and development of drug resistance in ovarian cancer.

Summary POLR3G is expressed at high levels in human pluripotent stem cells (hPSCs) and is required for maintenance of stem cell state through mechanisms not known in detail. To explore how POLR3G regulates stem cell state, we carried out deep-sequencing analysis of polyA+ and smallRNA transcriptomes present in hPSCs and regulated in POLR3G-dependent manner. Our data reveal that POLR3G regulates a specific subset of the hPSC transcriptome, including multiple transcript types, such as protein-coding genes, long intervening non-coding RNAs, microRNAs and small nucleolar RNAs, and affects RNA splicing. The primary function of POLR3G is in the maintenance rather than repression of transcription. The majority of POLR3G polyA+ transcriptome is regulated during differentiation, and the key pluripotency factors bind to the promoters of at least 30% of the POLR3G-regulated transcripts. Among the direct targets of POLR3G, POLG is potentially important in sustaining stem cell status in a POLR3G-dependent manner., Highlights • A specific subset of transcriptome is expressed in POLR3G-dependent manner in hPSCs • The primary function of POLR3G is in the maintenance of transcription • mtDNA polymerase gamma (POLG) is among the direct targets of POLR3G, In this article, Lund and colleagues show that RNA polymerase III subunit G (POLR3G) regulates a specific subset of transcriptome in human pluripotent stem cells. The primary function of POLR3G is in the maintenance of transcription. Among the direct targets of POLR3G is mtDNA polymerase γ, with potential importance in maintenance of stem cell state in a POLR3G-dependent manner.